spectralbranding/prism-c-choice

PRISM-C — Choice-Perception Gap Campaign

Complete, reproducible campaign data for PRISM-C, a pre-registered instrument that measures the gap between what AI observers say they perceive of brands (stated eight-dimension readings) and what they choose when the same brands compete in a concrete choice scenario — with the gap tested against the instrument's own choice-elicitation operator floor and a pre-registered mechanism battery fixed before any data collection.

• Paper (concept DOI): https://doi.org/10.5281/zenodo.21128342
• Code + machine-readable paper bundle: https://github.com/spectralbranding/sbt-papers/tree/main/prism-c-choice
• Campaign: 40 choice scenarios x 8 counterbalanced arrangements x 4 chooser families + stated readings for the full 40-brand bank; 2,981 logged model API calls (July 2026), frozen protocol, twin negative control, dominating-option positive control, mechanical operator-exclusion rule.

Structure

protocol/   PL0 preregistration (incl. pre-collection amendments), PL1 config,
            PL2 scenario bank, mechanisms (fixed ex ante), instrument scaffold
records/    parsed stated readings + choice records (JSONL), pilot report,
            PL4 estimator output
logs/       PL3 append-only call logs — one JSON row per model API call
            (prompts, parameters, responses, token usage)

Every reported number reproduces from records/ + the published seeded estimator in the GitHub mirror (fixed seed 20260702) without API keys.

Headline result

All three pre-registered hypotheses supported: choices diverge from stated-perception predictions on .633 of trials against a choice operator floor of .044 (S/N 14.5); choice-time dimension weights differ from stated weights with sign reversals (Experiential/Cultural proximity predicts NOT chosen); position effects are null on real brands while the twin negative control shows strong first-position rates — position bias appears only when perception is absent.

Citation

Zharnikov, D. (2026). From Stated Perception to Revealed Choice: A Pre-Registered Instrument for the Choice-Perception Gap in AI Brand Perception (PRISM-C). https://doi.org/10.5281/zenodo.21128342

License

CC BY 4.0 (data, this card). Companion code: MIT (GitHub mirror).